Problem with loading texture
Dr-Xiaogaren opened this issue · comments
Thanks for open-sourcing such a solid and excellent work. But I found the following problem in test:
When I tried to test the trained model on the Matterport3D dataset, I found that the texture of the scene mesh could not be loaded correctly. The RGB images acquired by the robots are completely black without color.
The scene used for testing is named E9uDoFAP3SH
picked from mp3d-b.scenes
. The data format of the scene is like this:
(base) [/workspace/NeuralCoMapping/venv/lib/python3.6/site-packages/gibson2/data/g_dataset/E9uDoFAP3SH]$ ls
combined.mtl floor_0.yaml floor_1.yaml floor_trav_0.png floor_trav_1.png ins_map.png mesh_z_up.obj.mtl texture_combined.jpg
floor_0.png floor_1.png floor_trav_0.p floor_trav_1.p floors.txt mesh_z_up.obj sem_map.png
I also did not modify the config file basic.yaml
. Although the lack of texture does not affect the acquisition of depth information, it would be better if a full RGB image could be acquired.
I would be very grateful if you could provide some ideas
It works fine in my environment. I guess you probably didn't set the vis_type parameter, whose default value 0 means "no visualization".
We disable RGB rendering by default for higher FPS. And I will add more notes about visualization in README soon :)
If you did set the vis_type to a non-zero value, plz let me know and I'll check further.